Replicate Scale a Deployment and Run a Prediction

arazzo: 1.0.1
info:
  title: Replicate Scale a Deployment and Run a Prediction
  summary: Read a deployment, update its version and instance bounds, then run a prediction through it.
  description: >-
    Operating an existing deployment often means rolling it to a new model
    version and adjusting its autoscaling before sending traffic. This workflow
    reads the current deployment, patches it with a new version and instance
    bounds, then runs a prediction through the updated deployment and polls it to
    completion. Every step spells out its request inline so the flow can be read
    and executed without opening the underlying OpenAPI description.
  version: 1.0.0
sourceDescriptions:
- name: deploymentsApi
  url: ../openapi/replicate-deployments-api-openapi.yml
  type: openapi
- name: predictionsApi
  url: ../openapi/replicate-predictions-api-openapi.yml
  type: openapi
workflows:
- workflowId: scale-deployment-and-predict
  summary: Update a deployment's version and scaling, then run a prediction through it.
  description: >-
    Reads the existing deployment, patches its version and min/max instances,
    submits a prediction through the updated deployment, and polls until the
    prediction reaches a terminal status.
  inputs:
    type: object
    required:
    - apiToken
    - deploymentOwner
    - deploymentName
    - version
    - minInstances
    - maxInstances
    - input
    properties:
      apiToken:
        type: string
        description: Replicate API token used as a Bearer credential.
      deploymentOwner:
        type: string
        description: The owner of the deployment to update.
      deploymentName:
        type: string
        description: The name of the deployment to update.
      version:
        type: string
        description: The model version ID to deploy.
      minInstances:
        type: integer
        description: The minimum number of instances for scaling.
      maxInstances:
        type: integer
        description: The maximum number of instances for scaling.
      input:
        type: object
        description: The model's input as a JSON object for the prediction.
  steps:
  - stepId: getDeployment
    description: >-
      Read the current deployment release so its existing configuration is known
      before updating it.
    operationId: deployments.get
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.apiToken
    - name: deployment_owner
      in: path
      value: $inputs.deploymentOwner
    - name: deployment_name
      in: path
      value: $inputs.deploymentName
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      currentVersion: $response.body#/current_release/version
      currentNumber: $response.body#/current_release/number
  - stepId: updateDeployment
    description: >-
      Patch the deployment with the new version and instance bounds. This
      increments the release number.
    operationId: deployments.update
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.apiToken
    - name: deployment_owner
      in: path
      value: $inputs.deploymentOwner
    - name: deployment_name
      in: path
      value: $inputs.deploymentName
    requestBody:
      contentType: application/json
      payload:
        version: $inputs.version
        min_instances: $inputs.minInstances
        max_instances: $inputs.maxInstances
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      newVersion: $response.body#/current_release/version
      newNumber: $response.body#/current_release/number
  - stepId: createDeploymentPrediction
    description: >-
      Run a prediction through the updated deployment.
    operationId: deployments.predictions.create
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.apiToken
    - name: deployment_owner
      in: path
      value: $inputs.deploymentOwner
    - name: deployment_name
      in: path
      value: $inputs.deploymentName
    requestBody:
      contentType: application/json
      payload:
        input: $inputs.input
    successCriteria:
    - condition: $statusCode == 201
    outputs:
      predictionId: $response.body#/id
  - stepId: getPrediction
    description: >-
      Retrieve the prediction state, repeating via the retry branch until the
      prediction reaches a terminal status.
    operationId: predictions.get
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.apiToken
    - name: prediction_id
      in: path
      value: $steps.createDeploymentPrediction.outputs.predictionId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      status: $response.body#/status
      output: $response.body#/output
    onSuccess:
    - name: predictionDone
      type: end
      criteria:
      - context: $response.body
        condition: $.status == "succeeded" || $.status == "failed" || $.status == "canceled"
        type: jsonpath
    - name: keepPolling
      type: retry
      retryAfter: 2
      retryLimit: 60
      stepId: getPrediction
      criteria:
      - context: $response.body
        condition: $.status == "starting" || $.status == "processing"
        type: jsonpath
  outputs:
    newReleaseNumber: $steps.updateDeployment.outputs.newNumber
    predictionId: $steps.createDeploymentPrediction.outputs.predictionId
    status: $steps.getPrediction.outputs.status
    output: $steps.getPrediction.outputs.output